禽兽的忏悔-_-
问题描述
我想从这个例子开始尝试使用 multiprocessing 的不同方式:
I wanted to try different ways of using multiprocessing starting with this example:
$ cat multi_bad.py
import multiprocessing as mp
from time import sleep
from random import randint
def f(l, t):
# sleep(30)
return sum(x < t for x in l)
if __name__ == '__main__':
l = [randint(1, 1000) for _ in range(25000)]
t = [randint(1, 1000) for _ in range(4)]
# sleep(15)
pool = mp.Pool(processes=4)
result = pool.starmap_async(f, [(l, x) for x in t])
print(result.get())
这里，l 是一个列表，当产生 4 个进程时会被复制 4 次.为了避免这种情况，文档页面提供了使用队列、共享数组或使用 multiprocessing.Manager 创建的代理对象.对于最后一个，我改变了l的定义:
Here, l is a list that gets copied 4 times when 4 processes are spawned. To avoid that, the documentation page offers using queues, shared arrays or proxy objects created using multiprocessing.Manager. For the last one, I changed the definition of l:
$ diff multi_bad.py multi_good.py 10c10,11 < l = [randint(1, 1000) for _ in range(25000)] --- > man = mp.Manager() > l = man.list([randint(1, 1000) for _ in range(25000)])
结果看起来仍然正确，但是执行时间急剧增加，以至于我认为我做错了什么:
The results still look correct, but the execution time has increased so dramatically that I think I'm doing something wrong:
$ time python multi_bad.py [17867, 11103, 2021, 17918] real 0m0.247s user 0m0.183s sys 0m0.010s $ time python multi_good.py [3609, 20277, 7799, 24262] real 0m15.108s user 0m28.092s sys 0m6.320s
文档确实说这种方式比共享数组慢，但这感觉不对.我也不确定如何对此进行分析以获取有关正在发生的事情的更多信息.我错过了什么吗?
The docs do say that this way is slower than shared arrays, but this just feels wrong. I'm also not sure how I can profile this to get more information on what's going on. Am I missing something?
附:使用共享数组，我得到的时间低于 0.25 秒.
P.S. With shared arrays I get times below 0.25s.
附言这是在 Linux 和 Python 3.3 上.
P.P.S. This is on Linux and Python 3.3.
推荐答案
Linux 使用 copy-当子进程被 os.forked 时，on-write.演示:
Linux uses copy-on-write when subprocesses are os.forked. To demonstrate:
import multiprocessing as mp
import numpy as np
import logging
import os
logger = mp.log_to_stderr(logging.WARNING)
def free_memory():
total = 0
with open('/proc/meminfo', 'r') as f:
for line in f:
line = line.strip()
if any(line.startswith(field) for field in ('MemFree', 'Buffers', 'Cached')):
field, amount, unit = line.split()
amount = int(amount)
if unit != 'kB':
raise ValueError(
'Unknown unit {u!r} in /proc/meminfo'.format(u = unit))
total += amount
return total
def worker(i):
x = data[i,:].sum() # Exercise access to data
logger.warn('Free memory: {m}'.format(m = free_memory()))
def main():
procs = [mp.Process(target = worker, args = (i, )) for i in range(4)]
for proc in procs:
proc.start()
for proc in procs:
proc.join()
logger.warn('Initial free: {m}'.format(m = free_memory()))
N = 15000
data = np.ones((N,N))
logger.warn('After allocating data: {m}'.format(m = free_memory()))
if __name__ == '__main__':
main()
产生了
[WARNING/MainProcess] Initial free: 2522340 [WARNING/MainProcess] After allocating data: 763248 [WARNING/Process-1] Free memory: 760852 [WARNING/Process-2] Free memory: 757652 [WARNING/Process-3] Free memory: 757264 [WARNING/Process-4] Free memory: 756760
这表明最初大约有 2.5GB 的可用内存.在分配 15000x15000 的 float64 数组后，有 763248 KB 可用空间.这大致是有道理的，因为 15000**2*8 字节 = 1.8GB 并且内存下降，2.5GB - 0.763248GB 也大约是 1.8GB.
This shows that initially there was roughly 2.5GB of free memory. After allocating a 15000x15000 array of float64s, there was 763248 KB free. This roughly makes sense since 15000**2*8 bytes = 1.8GB and the drop in memory, 2.5GB - 0.763248GB is also roughly 1.8GB.
现在每个进程生成后，可用内存再次报告为 ~750MB.可用内存没有显着减少，因此我认为系统必须使用写时复制.
Now after each process is spawned, the free memory is again reported to be ~750MB. There is no significant decrease in free memory, so I conclude the system must be using copy-on-write.
结论:如果您不需要修改数据，在 __main__ 模块的全局级别定义它是一种方便且(至少在 Linux 上)内存友好的方式来共享它子进程.
Conclusion: If you do not need to modify the data, defining it at the global level of the __main__ module is a convenient and (at least on Linux) memory-friendly way to share it among subprocesses.
